基于神经网络的绝缘栅双极型晶体管开关损耗预测  

Insulated gate bipolar transistor switching loss prediction based on neural network

作　　者：王长华 李祥雄 梁顺发 陈荣东 WANG Changhua;LI Xiangxiong;LIANG Shunfa;CHEN Rongdong(SUNTEN Electrical Equipment Co.,Ltd,Foshan,Guangdong 528300)

机构地区：[1]顺特电气设备有限公司,广东佛山528300

出　　处：《电气技术》2025年第3期42-48,共7页Electrical Engineering

摘　　要：针对级联储能应用领域大量绝缘栅双极型晶体管(IGBT)的开关损耗难以准确在线测量的问题,引入误差反向传播神经网络,建立IGBT开关损耗预测模型。首先采用级联H桥功率模块搭建开关损耗动态测试系统,通过调整直流母线电压、交流电流及冷却液温度,获得大量测试数据;然后将影响IGBT开关损耗的3个主要因素——集射极电压、集电极电流、结温作为预测模型的输入,采用粒子群优化算法优化开关损耗预测模型的初始权值和阈值,以提升预测开通损耗、关断损耗及二极管反向关断损耗的准确度并加速学习规律的收敛;最后与随机给定初始权值及阈值的开关损耗预测模型进行对比分析。结果表明,引入粒子群优化算法所建立的开关损耗预测模型的预测准确度更高,针对50组随机验证数据的最大百分误差为3.3%。Aiming at the disadvantages that numerous insulated gate bipolar transistor(IGBT)switching loss are difficult to accurately measure online in the cascaded energy storage application area,switching loss prediction model is established based on the error back propagation neural network.Firstly,dynamic test system of switching loss is built with cascaded H bridge power module,the massive switching loss data is obtained with changing the direct current bus voltage,alternating current and coolant temperature of power module.3 main factors including collector-emitter voltage,collector current and device junction temperature are taken as the input of IGBT switching loss prediction model.The particle swarm optimization is used to optimize the initial weight and threshold of prediction model,improving prediction accuracy and accelerating the convergence of learning laws.The optimized performance of this model is compared and analyzed with the prediction model that the initial weight and threshold are given randomly.The results show that the prediction accuracy of the model proposed in this paper is higher.The maximum percentage error for 50 sets of random validation data is 3.3%.

关 键 词：绝缘栅双极型晶体管(IGBT) 开关损耗预测 神经网络 粒子群优化算法 

分 类 号：TN322.8[电子电信—物理电子学] TP183[自动化与计算机技术—控制理论与控制工程]